Real-Time Vision-Controlled Magnetic XY Arena
Closed-loop camera tracking and motorized XY magnet stage moving an autonomous stimulus inside a chamber to measure visually mediated responses in walking fruit flies (Current Biology, 2012).
SYS.00 // APPLIED AI · GENERATIVE, AGENTIC & PHYSICAL SYSTEMS
CALIFORNIA, USA · @DATASITH
cisco zabala @datasith
Applied AI architect, former robotics engineer, and perception researcher building generative, agentic, and physical AI systems from whiteboard to production. Focused on how agents use tools, reason over governed data, and work alongside people on hard real-world problems.
ARCH_01 // OPEN-SOURCE REFERENCE BLUEPRINTS
APACHE 2.0 · TERRAFORM + ADK
GCP Agentic AI Samples: Cross-Platform A2A Interop & Continuous Evaluation Flywheel
Production-grade, runnable multi-agent architectures provisioned end-to-end with Terraform. Sample 01 wires bidirectional 3P SaaS-to-Cloud Agent Runtime interoperability over A2A (HTTP+JSON), OIDC Workload Identity Federation, Agent Registry, and dual-mode Agent Gateway. Sample 02 implements a continuous ADK evaluation flywheel spanning local trajectory checks, Metric Registry custom evaluators, synthetic user simulation, loss-cluster root-cause analysis, and OpenTelemetry GenAI v1.39+ ship gates.
INTEROP PROTOCOL
A2A · OIDC WIF · MCP
QUALITY GATE
ADK Eval · OTel v1.39+
PROVISIONING
Modular Terraform
RESEARCH_02 // CLOSED-LOOP PERCEPTION & KINEMATICS
CURRENT BIOLOGY · IEEE TBME · JEB
Animal-Robot Visual Interaction & 3D High-Speed Kinematic Tracking
Before foundation models and cloud agents, built real-time closed-loop perception and robotic instrumentation for studying biological sensorimotor control: a vision-guided magnetic XY arena (Flyatar) interacting with walking Drosophila in real time, and a self-triggered multi-camera high-speed motion-capture rig (Flytrax) reconstructing 3D wing and body flight maneuvers.
CONTROL LOOP
Real-Time XY Vision
CAPTURE RIG
3D High-Speed Video
PUBLICATIONS
03 Journal Papers
CAPABILITIES // FULL-STACK APPLIED AI
Technical Architecture Domains
Spanning cloud-scale multi-agent systems, self-supervised vision foundation models, and embedded edge hardware.
STACK_01
Agentic & Generative AI
Multi-agent orchestration (ADK, A2A, MCP), governed RAG, trajectory evaluation flywheels, synthetic user simulation, vLLM, and runtime guardrails.
ADK · A2A · MCP
STACK_02
Vision & Foundation Models
PyTorch, self-supervised pre-training (MAE, DINO), Vision Transformers, multi-GPU distributed training, TensorRT, ONNX, and TFLite edge quantization.
PyTorch · ViT
STACK_03
Cloud & MLOps Infrastructure
Multi-cloud AI architectures across Google Cloud and AWS, Terraform IaC, Docker, Kubernetes, OpenTelemetry GenAI observability, and CI/CD ship gates.
GCP · AWS · IaC
STACK_04
Physical AI, Edge & Robotics
Vision-language-action (VLA) pipelines, ROS/ROS2, high-speed multi-camera capture rigs, embedded C/C++ firmware, custom PCB instrumentation, and IoT.
ROS2 · Edge VLA
STACK_05
Languages & Data Systems
Python, C/C++, SQL, TypeScript/JavaScript, vector databases, streaming telemetry pipelines, and deterministic statistical evaluation.
Python · C++ · SQL
